ray/rllib/tuned_examples/halfcheetah-ppo.yaml

halfcheetah-ppo:
    env: HalfCheetah-v2
    run: PPO
    stop:
        episode_reward_mean: 9800
        time_total_s: 10800           
    config:
        gamma: 0.99
        lambda: 0.95
        kl_coeff: 1.0
        num_sgd_iter: 32
        lr: .0003
        vf_loss_coeff: 0.5
        clip_param: 0.2
        sgd_minibatch_size: 4096
        train_batch_size: 65536
        num_workers: 16
        num_gpus: 1
        grad_clip: 0.5
        num_envs_per_worker: 
            grid_search: [16, 32]
        batch_mode: truncate_episodes
        observation_filter: MeanStdFilter
Optimal PPO Configs (10k reward in 1 hr) + PPO grad clipping implemented (#3934) 2019-02-02 22:10:58 -08:00			`halfcheetah-ppo:`
			`env: HalfCheetah-v2`
			`run: PPO`
			`stop:`
			`episode_reward_mean: 9800`
			`time_total_s: 10800`
			`config:`
			`gamma: 0.99`
			`lambda: 0.95`
			`kl_coeff: 1.0`
			`num_sgd_iter: 32`
			`lr: .0003`
			`vf_loss_coeff: 0.5`
			`clip_param: 0.2`
			`sgd_minibatch_size: 4096`
			`train_batch_size: 65536`
			`num_workers: 16`
			`num_gpus: 1`
			`grad_clip: 0.5`
			`num_envs_per_worker:`
			`grid_search: [16, 32]`
			`batch_mode: truncate_episodes`
[rllib] Set PPO observation filter to NoFilter by default (#4191) 2019-03-01 13:19:33 -08:00			`observation_filter: MeanStdFilter`